Piezoelectric actuators have been widely applied in micro-positioning devices. The hysteresis nonlinearity of piezoelectric materials degrades the control accuracy. To improve the system performance, modern controllers are designed with a mathematical model which can describe the hysteresis behavior. Preisach model is the most popular model in hysteresis modeling, and it can precisely describe the hysteresis of piezoelectric actuators. But classic Preisach model needs a large quantity of weight parameters to achieve more accuracy, which complicates the model identification and controller design. Neural network has a lot of successful applications in system identification, and its weight parameters are easy to be identified with available training algorithm, but it cannot describe multi-valued mapping of hysteresis. A neural-Preisach model is proposed for modeling and control of piezoelectric actuators. The neural-Preisach model inherits the advantages of Preisach model and neural network, which can describe the hysteresis and update parameters by training algorithm. An inverse controller was designed with the inverse neural-Preisach model, and then experiments of tracking control were performed to validate the effectiveness of the neural-Preisach model. The maximal error is reduced by 74.50%, compare in case without control. This indicates that control accuracy with hysteresis compensation is greatly improved compared to that without hysteresis compensation.